FEPP: SOFTWARE RISK PREDICTION IN REQUIREMENTS ENGINEERING USING RULE EXTRACTION AND MULTI-CLASS INTEGRATION

Abstract

Requirements Engineering (RE) is a major software development phase in which vague, incomplete, or inconsistent requirements pose major risks that delay the project, cause cost escalation, and result in project failure. Existing risk prediction models are mostly based on binary classification and are not flexible enough to cope with multi-dimensional risks. This work suggests a Feature-Enriched Prediction Paradigm (FEPP) that blends rule extraction methods with multi-class classification methodologies to improve software risk prediction during the RE phase. FEPP employs Association Rule Mining (ARM) and Fuzzy Logic in extracting dynamic rules from past software project experiences and projecting future risk patterns. Multi-class models such as Random Forest, XGBoost, and Gradient Boosting are being combined in the form of Voting Classifier Mechanism (VCM) in order to enhance prediction accuracy. Findings based on large-scale experimentation on standard RE datasets reveal that FEPP increases risk prediction accuracy by 12-15% compared to existing models with effective early risk mitigation and better project outcomes.